hughie

Tool Sharing#

FastSAM is a CNN Segment Anything model trained using only 2% of the SA-1B dataset. FastSAM achieves performance comparable to the SAM method at 50 times the speed.

Features#

FastSAM is officially introduced as being 50 times faster than SAM while achieving the same performance. The model is based on the YoloV8 architecture and has much lower training costs compared to SAM.

Deployment#

• Installation

  Copy

  git clone https://github.com/CASIA-IVA-Lab/FastSAM.git && cd FastSAM
  pip install git+https://github.com/openai/CLIP.git
  
  conda create -n FastSAM python=3.9
  conda activate FastSAM
  
  pip install -r requirements.txt
  

• Usage

  Copy

  # Everything mode
  python Inference.py --model_path ./weights/FastSAM.pt --img_path ./images/dogs.jpg
  
  # Text prompt segmentation with CLIP assistance
  python Inference.py --model_path ./weights/FastSAM.pt --img_path ./images/dogs.jpg  --text_prompt "the yellow dog"
  
  # Box prompt mode
  python Inference.py --model_path ./weights/FastSAM.pt --img_path ./images/dogs.jpg --box_prompt "[570,200,230,400]"
  
  # Click mode
  python Inference.py --model_path ./weights/FastSAM.pt --img_path ./images/dogs.jpg  --point_prompt "[[520,360],[620,300]]" --point_label "[1,0]"
  

Platform#

Official demo

Disclaimer#

This article is solely for the purpose of sharing tools.

This article is synchronized with HBlog.